Extending Geant4 Parallelism with External Libraries (MPI, TBB) and Its Use on HPC Resources

TL;DR

This paper discusses enhancements to Geant4's parallel processing capabilities, integrating external libraries like MPI and TBB, to improve scalability on high-performance computing resources.

Contribution

It introduces new methods for integrating MPI and TBB with Geant4, enabling hybrid parallelism across nodes and cores for large-scale HPC applications.

Findings

Successful implementation of MPI/MT hybrid parallelism in Geant4.

Enhanced scalability on HPC clusters with multi-node and multi-core configurations.

Improved support for external libraries to facilitate large-scale simulations.

Abstract

With Geant4 Version 10.0, released in December 2013, one of the most widely used Monte-Carlo codes has been ported to take full advantage of multi- and many-core CPUs thanks to the introduction of event-level parallelism via multithreading. In this paper we review recent developments to allow for a better integration of parallel Geant4 jobs with external libraries. We have chosen to develop examples using the popular Intel Threading Building Block (for short TBB) as an alternative parallelization approach to the native Geant4 POSIX. To simplify the scaling of a Geant4 application across nodes on a cluster we are improving the support of MPI in Geant4. In particular it is now possible to run an hybrid MPI/MT application that uses MPI to scale across nodes and MT to scale across cores. %The recent developments allow users to easily implement parallel application resources that scale on a very large number of nodes and cores typical of HPC resources.